Creating a Veteran's specific risk model to improve lung cancer screening

目前的肺癌筛查资格指南是在平民人群中制定的， 大多数退伍军人谁发展肺癌。该指南包括50-80岁的重度吸烟者， 或以上包年的历史，谁要么目前吸烟或戒烟在过去的15年。这些标准仅 在平民和退伍军人中捕获20-35%的肺癌。此外，退伍军人遭受 肺癌的发病率高于美国其他人口，吸烟更多，并且具有独特的 暴露于已知的肺癌病因，包括橙子剂、石棉、柴油烟雾、电离辐射 和露天焚烧坑碳氢化合物。退伍军人也有肺癌的其他风险因素，如种族，低 社会经济状况、既往癌症、HIV、类风湿性关节炎和慢性阻塞性肺疾病史 慢性阻塞性肺疾病（COPD），其中每一种都显示出增加肺癌风险。其他，特定人群 模型有效地识别出可能从筛查中获益的风险亚组，但这些模型都没有 在退伍军人中得到验证，没有人考虑退伍军人的独特风险。个性化和退伍军人专用 该模型增加了与服务相关的肺癌风险，并导致识别可能 需要从肺癌筛查中获益。本提案的目的是将联合收割机 和退伍军人特异性肺癌危险因素纳入退伍军人肺癌筛查资格模型。我们 总体假设是，服务历史和新的风险因素可以用于退伍军人特异性肺癌 风险模型，以扩大可能受益于肺癌筛查的人群。这种努力改善 通过筛查早期发现肺癌，退伍军人的健康有两个目的。 在目标1中，我们将定义和发现与肺癌风险增加相关的新表型， 退伍军人，包括纵向临床和军事服务特定的风险。我们将生成一个 来自所有在VA接受护理的退伍军人的一组全面的纵向肺癌风险因素 设施在过去的十年里。我们将使用链接的国防部服务和VA电子健康记录 (EHR)数据，以确定与服务有关的暴露和肺癌风险因素。利用人工智能，我们 将从临床笔记和放射学报告中挖掘非结构化文本数据，以发现新的数据模式 （表型）有助于预测未来的肺癌诊断。我们假设我们能准确地确定 风险变量用于当前的资格模型，并发现了一组新的退伍军人特异性表型 与肺癌风险相关。在目标2中，我们将建立退伍军人特异性肺癌筛查模型 并将其与现有的筛选资格标准和模型进行比较。我们将使用标准的 肺癌风险变量、兵役特异性风险因素和新发现的EHR肺癌风险 表型来开发肺癌筛查模型。该模型的变量将包括丰富的马赛克 静态和时变指标（吸烟行为、实验室值、肺功能等），肺癌风险 EHR表型（COPD、HIV等），和服务特定风险（橙子剂、石棉等）。我们将 将我们的新模型与现有的肺癌筛查指南、巴赫、利物浦肺项目和 PLCO筛选资格模型。我们假设退伍军人特定模型将识别出更多的风险 与目前的筛选资格模型相比，个人具有更高的准确性和校准性。 与全国公认的领导者在肺癌，信息学，VA数据使用，机器学习，流行病学， 和生物统计学，我们处于独特的地位，以实现这些目标。在完成本提案后， 将开发退伍军人特定模型，并与现有的肺癌筛查资格模型进行比较， 高危退伍军人

Current lung cancer screening eligibility guidelines were developed in a civilian population and miss the majority of Veterans who develop lung cancer. The guidelines include 50-80 year old heavy smokers, with a 20 or more pack years history, who either currently smoke or quit within the last 15 years. These criteria only capture 20-35% of lung cancers in the civilian population and Veterans. Furthermore, Veterans suffer from lung cancer at higher rates than the rest of the United States population, smoke more, and have unique exposures to known causes of lung cancer including Agent Orange, asbestos, diesel fumes, ionizing radiation and Open Burn Pit hydrocarbons. Veterans also have additional risk factors for lung cancer such as race, low socio-economic status, previous history of cancer, HIV, rheumatoid arthritis and chronic obstructive pulmonary disease (COPD) each of which have been shown to increase lung cancer risk. Other, population specific models effectively identify at risk subgroups who may benefit from screening, but none of these models have been validated in Veterans and none consider Veterans’ unique risks. A personalized and Veteran-specific model that adds service-related lung cancer risks and leads to the identification of high-risk groups that may benefit from lung cancer screening is needed. The objective of this proposal is to combine general population and Veteran-specific lung cancer risk factors into a Veteran's lung cancer screening eligibility model. Our overall hypothesis is that service histories and novel risk factors can be used in a Veteran-specific lung cancer risk model to broaden the population who may benefit from lung cancer screening. This effort to improve Veterans’ health through the early detection of lung cancer with screening has two aims. In Aim 1 we will define and discover novel phenotypes associated with increased lung cancer risk in Veterans that include longitudinal clinical and military service-specific exposures. We will generate a comprehensive, longitudinal set of lung cancer risk factors from all Veterans who have received care at a VA facility in the last decade. We will use linked Department of Defense service and VA Electronic Health Record (EHR) data to identify service-related exposures and lung cancer risk factors. Using artificial intelligence, we will mine unstructured text data from clinical notes radiological reports to discover novel data pattern (phenotypes) that help predict future lung cancer diagnosis. We hypothesize that we will accurately determine risk variables used in current eligibility models and discover a set of novel Veteran-specific phenotypes associated with lung cancer risk. In Aim 2 we will build a Veteran-specific lung cancer screening model and compare it to existing screening eligibility criteria and models. We will use a combination of standard lung cancer risk variables, military service-specific risk factors and novel discovered EHR lung cancer risk phenotypes to develop a lung cancer screening model. The variables for this model will include a rich mosaic of static and time varying metrics (smoking behavior, lab values, pulmonary function, etc.), lung cancer risk EHR phenotypes (COPD, HIV, etc.), and service-specific risks (Agent Orange, asbestos, etc.). We will compare our new model to the existing lung cancer screening guidelines, the Bach, Liverpool Lung Project and PLCO screening eligibility models. We hypothesize that a Veteran-specific model will identify more at-risk individuals with greater accuracy and calibration compared to current screening eligibility models. With nationally recognized leaders in lung cancer, informatics, VA data use, machine learning, epidemiology, and biostatistics, we are uniquely positioned to accomplish these goals. At the completion of this proposal, a Veteran-specific model will be developed and compared to existing lung cancer screening eligibility models for at-risk Veterans.